🚀 go-pugleaf

On Thu, 13 Aug 1992 15:30:01 GMT, NEWS@UCF1VM.BITNET said:
(Can't somebody fix the blasted gateway so that half the original
messages don't lose their headers?)

> (...)
> Pipelining and alignment
> and god knows what else all affect instruction execution rates,
>
> Anyway, he clocked the following instructions at the indicated MIPS:
>
>    ICM   0,15,W1             10.4287
>    L     0,W1                41.8938
>    LTR   0,1                 41.7416
>
> So it appears that -- at least on a 3081K -- the L/LTR sequence is
> roughly twice as fast as the ICM.

That's not a valid conclusion.  You can't add execution times of
instructions to get the execution time of the combination, exactly for
the reason you mention.  When we got a 3090-200 at my previous
employer, the first thing that I did was try to compare its speed to
our 3084Qx.  (Yep, that's a lower case x, which was halfway between a
3084Q and a 3084QX.)  I ran various tiny programs, like a BCTR loop to
clear a register, which I think ran at 40 MIPS.  I kept adding
instructions into the loop.  If I remember right, a loop consisting of
LR 1,2; BCTR 3,14 was the same speed as just the BCTR.  In fact, I
could add carefully chosen instructions without really affecting the
CPU time, so I got some wild MIPS results.  Most floating point
instructions also took zero time.  I had almost 10 instructions in the
loop and it still ran in about the same time as just the BCTR loop,
until I added a floating point multiply, at which point the CPU time
shot up.

So, the only correct way to determine whether L;LTR is faster than
ICM 15 is to time it on your machine.  But the CPU time used in this
discussion is probably much higher than any potential savings from
knowing the answer.

/Leonard